The Politics of AI - II: Unchaining the Future

"Thus in depending on the future of invention to extricate us from the situations into which the squandering of our natural resources has brought us we are manifesting our national love for gambling and our national worship of the gambler, but in circumstances under which no intelligent gambler would care to make a bet. Whatever skills your successful poker player must have, he must at the very least know the values of his hands. In this gamble on the future of inventions, nobody knows the value of a hand.” Norbert Wiener, The Human Use of Human Beings, 1950

Life in the 21st century is reader supported, please subscribe to make life sustainable.

As we’ve never had understanding of any politics of technology, compute has followed the same development of previous technologies. Focus falls exclusively on the technology itself with complete disregard of the technology’s larger social impacts, along with totally ignoring the greater environmental consequences. Wiener claimed information to be neither matter or energy, however information only manifests itself through energy, matter, or both. With compute technologies, information manifests through transistors, a digital process producing 1s and zeros by opening and closing transistors with electricity. The more transistors, the more electricity needed. With AI, in order to run what is more than anything else massive quantities of compute power in a centralized network architecture, an enormous increase in electricity generation is required.

Over the last few years, for the first time, we finally began to take into account the environmental ramifications of a century of electricity generation. This has now all been swept to the side as Tech's faux-prophets call for a doubling or even tripling of generation to meet the next level of their poorly imagined manufactured nirvana.

The MIT Technology Review writes in less than five years Google doubled their electricity consumption. Such large scale growing demand creates a mad scramble for electricity generation both within the industry and society as a whole. S&P Global writes, “Recent projections have data center energy consumption doubling to nearly 800 TWh by 2030 — more than triple the amount of electricity California consumed in 2023.”

Amusingly, Google recently announced they’ve signed an agreement for fusion power, a technology that doesn’t exist. Fusion electricity generation has been twenty years away for 75 years now. Of course this is well within Tech's tradition of promising things they can’t provide. With no irony, an AI technologist writes, “The tech sector is aware that AI emissions put its climate commitments in danger. Both Microsoft and Google no longer seem to be meeting their net zero targets. So far, however, no Big Tech company has agreed to use the methodology (measuring energy use) to test its own AI models.” While Nvidia, not to be out done, promotes, “Climate in a Bottle: A generative foundation model for the kilometer-scale atmosphere.” Let’s just call it stupid, but nothing better demonstrates how Tech moves with no perspective other than that of the technology they are developing. The technology leads itself and us with no concern for anything else.

The industry is calling for a whole new fleet of nuclear plants, a technology, much like the transistor, we have yet to in anyway come to reckon with its political, cultural, or environmental impacts. The last nuclear plant opened two years ago in Georgia was seven years late and $17 billion over budget. Microsoft, long the most ruthlessly practical of the entire Tech lot, announced they're restarting Three Mile Island nuclear facilities. It's no coincidence an industry seeking ever greater centralization of the control of information calls for the installation of the ultimate centralized electricity generation mechanism.

The splitting of the atom and the creation of the transistor were accomplished a few years apart. The question of mass destruction with nuclear weapons still, like a post-modern Sword of Damocles, hangs above humanity’s collective head. The unanswered and unsolved problems of nuclear powered electricity generation still remain. "A pretty elaborate process to boil water,” physicist Dr. Ferguson would quip.

The problems with nukes in short:

1) The waste remains for tens of thousand of years. It’s safe interment is still unsolved with 80 years worth of waste now stored at the plants that created them.

2) The waste creates fuel that can be refined to create nuclear weaponry, that's what the Manhattan Project's Oak Ridge reactor did.

3) Accidents, even if few and far between, can be quite costly.

4) It allows a very small priesthood to control electricity generation, not that that’s in anyway an exclusive problem of nukes in the electric industry, but that it's now specifically being promoted for the centralized control of information brings the problem to a whole other level.

Basically, these are all questions that need to be decided by society as a whole, not a few technologists and profiteers. If society as a whole wishes to build these plants, fully aware of all its implications, it is their decision to make. Shame on us all if we allow them to be built for this.

Finally, the NYT reports Wall Street wants to cash in: Wall St. Firms Are Buying Utilities to Tap Into the A.I. Boom. “BlackRock, the world’s largest asset manager, last year proposed buying Minnesota Power, a utility that owns several power plants and thousands of miles of power lines that could help technology companies secure energy for their data centers.” Who will pay for all this? Well, as they say, if you gotta ask… “US utilities plot big rise in electricity rates as data center demand booms.”

Electricity is the necessary energy component of manifesting this information revolution, water is a necessary matter aspect. Massive data centers using great amounts of energy require huge quantities of water for cooling. The Guardian reports TikTok’s plan to build a massive data center in Brazil in an already drought prone area. While in the US, the NYT reports, “After Meta broke ground on a $750 million data center on the edge of Newton County, Ga., the water taps in Beverly and Jeff Morris’s home went dry.”

Information technology is very much an industry. We have yet, not in anyway, come to terms with the environmental and social impacts of the industrialism that preceded it. All this feedback is in direct opposition to the abundance nonsense peddled by Tech – no one’s coding another planet.

A very important and too little understood fact about technological development is once society adopts a technology, that technology constrains future development around it. Technologies can define their environments to such an extent they come to be perceived as innate, too difficult to contemplate fundamentally altering without collapsing the systems created around it.

The easiest example of this is American car culture. Changing the American transportation system is to move away from the automobile, yet, the American imagination that can dream AI, can’t imagine a non-automobile centric-transportation system. Redesigning the infrastructure to be less car dependent is not simply a matter of technology, but requires a politics that takes on powerfully entrenched corporations, their government representatives, a very entrenched culture, and maybe most immovable, a public perception that this is simply how the world works. As the present information technology infrastructure develops, it will increasingly disallow alternatives, continuing the development of the recently established centralized information and communication infrastructure, and working to further centralize control.

From a political perspective, we see the same established forces – Wall Street, the utilities, and of course Tech itself – both controlling and most benefiting from AI. The entire process is directed by a handful of corporations, the largest and most significant being Microsoft, Amazon, and Google. From any historical perspective, the control of information and its communication are fundamental to defining homo sapiens interactions, our culture, our politics. With little exaggeration, under the present trajectory, what is being developed is one of the most tyrannically controlled societies in human history.

To gain any grasp what the industry is doing, we need to disregard the term intelligence. This is not meant to get in a discussion about what is or is not intelligence, which is both well beyond the scope of this brief piece and in many ways irrelevant. Any such discussion would need to start with the fact there is no genuinely accepted, objective, scientific definition of intelligence. That the definition of intelligence remains very much indistinct, any claim to any sort of artificial derivative should strike all as at very best problematic, though it does make a very fine, if completely ambiguous, marketing slogan. Aren’t those always the best?

At a very general level, there are some essential components to what we more or less agree is intelligence, for example memory. A good memory goes a long way representing most people’s perception of intelligence. Providing raw information that can be regurgitated at will, even if the remembered information is little understood but simply available, goes a long way in people’s assessment of intelligence. Memory, the immediate access to massive amounts of data, is essential for AI. Most of AI’s software and compute processes concern combing through massive banks of data. It is access to all this information, literally memory, that AI has most in common with what can be loosely defined as intelligence.

The capabilities developed to access all this information should in no way be underestimated, nor the mass amounts of data now electronically available, though certainly the question of the right to access and compensation for that information needs much greater discussion. The software process of retrieving and organizing provides the data context. Also not well known is much of this ordering, categorization, requires the direct human input of “thousands of low-cost workers in countries such as Kenya and the Philippines, who are typically paid less than $2 an hour to undertake the time-consuming task of annotating the huge datasets used to train AI models.” This FT article also claims these people are now being replaced “with highly paid industry specialists, in the latest push to build ‘smarter’ and more powerful models.” Good on the FT for putting quotes around smarter.

This labeling allows a faster software sorting, a searching of the data already provided context by human intellect. With AI, this software sorting is done repeatedly. The massive growth of compute power allows both the amount of data and its processing, its sorting, to be done at levels and speeds unavailable just a decade ago. Again, such massive quantities of data, no matter how well labeled, and its processing requires energy.

This artificial process is in so many ways different, for example in the physical components of the system or the massive amounts of energy required, that the comparison of whatever is perceived as human intelligence, provides us little understanding of the technology. It is more helpful to simply look what can be done with AI, most essentially, the automation of physical processes. The key here is repetition and certainly the debate about repetition as intelligence, both natural and artificial is one worth having. There is absolutely no doubt the technology is developing capabilities to take a giant leap in automating many repetitive aspects of society. How such automation fits into society is a political question for which we have no politics.

We can separate, somewhat arbitrarily, automation into two distinct categories. First is physical, such as manufacturing, checkouts at the store, and driving vehicles. Other aspects of automation can be considered more abstract than physical, or simply the manipulation of information and its communication, such as filling in forms and delivering the gathered information to the right destination. Or another example is demonstrated by enhanced search capabilities, for example Chatgpt, or any other such systems. Any belief you’re getting some sort of objective answer to your query, most especially in regards to questions inherently subjective, is both wrong thinking or Tech propaganda. Foremost, all AI is subjected to the data inputted, an aspect of all intelligence whether its natural or artificial. However, it is incredulous to believe, looking at the present internet and its profit function as an advertising platform, that you’re going to get objective information. As the industry knows nothing else, AI will definitely be profitized by selling results – caveat emptor.

In conceptualizing automation, the Cartesian idea of mind/body duality may finally be put to permanent rest. It is increasingly clear from our growing understanding of biology, the body is essential in defining the mind. With AI, the physical architecture is also determinative. The French AI firm Dataiku, funded by Google, has some essential thoughts on this. It should first be noted this firm is greatly representative of the current AI boom. Wikipedia reports they’ve raised almost a billion dollars in 12 years with their last round of debt valuing them at $3 billion, though reported profits remain slim. Good work if you can get it I suppose.

Nonetheless, with the money they’ve raised, they’ve done some good thinking on these matters. In a piece titled 3 Keys to a Modern Data Architecture Strategy Fit for Scaling AI, they advise to not over centralize. “When efforts to centralize have failed over and over (and over and over) again, the answer isn’t to double down on centralization, but that’s often the reality.” Yet, the entire Tech industry is in mad rush to do just that. They humorously add, “If IT owns the source of truth, but no one agrees with it, what use is that centralization?” This folks is the question we should all be asking.

Also Dataiku recommends “allowing people from around the organization to gather data themselves (even from multiple sources) and merge it without IT intervention.” This essential insight is not simply for business, but imperative for society as a whole in thinking about the processes of compute technology. In many ways, in vastly assorted different ways, people need to be kept in the process. It is the only way to have any hope of democratizing both the processes and more importantly society as a whole. Yet, we have little organizational understanding or capacities capable of doing this both in IT or in reforming our moribund political institutions. Instead we follow the siren songs of a handful of corporations claiming they know the future – centralization and people removal.

In developing an immediate politics of AI, the first step must be balancing the present monopoly control of the technologists with a much more diverse representation of society. Surprisingly, a simple first step was recently taken by the US Senate last week when they blocked the industry’s attempt to keep the states out of regulation, a small step in the right direction. A second step can be shutting-off the ridiculous attempt to massively growing electricity generation by building a new fleet of nuclear power plants. Focus on this can, again, be largely at the state and local levels, this is where utilities are regulated. Creating a politics for any sort of healthy AI means recreating democracy in America. It means restructuring government back to the local level and then networking the local with the understanding we need an active and engaged citizenry with access to political and government institutions where they can physically come together to talk, debate, and decide the future.

In “A is for Atom,” Alvin Weinberg noted “Energy is the ultimate raw material.” Yet, no energy is free. Tech has now made obvious their present configuration for future development comes with unmeetable energy demands. The planet has still to come to terms with the environmental and other societal problems incurred with the massive manipulation of energy utilized for the last two centuries of industrialization. If we can’t use all this information to make us smarter about energy use, it's a detrimentally limited intelligence, both artificial and natural.

Claims on the future, always somewhat or fully arbitrary, are now sold as ludicrously definitive politics. The future is contingent, getting there requires one foot after the next with an absolute understanding each step taken needs to be carefully assessed to where we are now and from where we came, only then can we make any intelligent queries on where to go next. A healthy democratic understanding of any of those perspectives requires more than a view from one technology or one small group of profiteers. It needs the input of the whole of society.

Life in the 21st century is reader supported, please subscribe and make life sustainable.